{
#endif
struct osd_iobuf *iobuf = bio->bi_private;
- int iter;
struct bio_vec *bvl;
/* CAVEAT EMPTOR: possibly in IRQ context
/* the check is outside of the cycle for performance reason -bzzz */
if (!bio_data_dir(bio)) {
- bio_for_each_segment_all(bvl, bio, iter) {
+ DECLARE_BVEC_ITER_ALL(iter_all);
+
+ bio_for_each_segment_all(bvl, bio, iter_all) {
if (likely(error == 0))
SetPageUptodate(bvl_to_page(bvl));
LASSERT(PageLocked(bvl_to_page(bvl)));
return bio_end_sector(bio) == sector ? 1 : 0;
}
+#if IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)
/*
* This function will change the data written, thus it should only be
* used when checking data integrity feature
static void bio_integrity_fault_inject(struct bio *bio)
{
struct bio_vec *bvec;
- int i;
+ DECLARE_BVEC_ITER_ALL(iter_all);
void *kaddr;
char *addr;
- bio_for_each_segment_all(bvec, bio, i) {
+ bio_for_each_segment_all(bvec, bio, iter_all) {
struct page *page = bvec->bv_page;
kaddr = kmap(page);
bip->bip_vec->bv_offset;
struct bio_vec *bv;
sector_t sector = bio_start_sector(bio);
- unsigned int i, sectors, total;
+ unsigned int sectors, total;
+ DECLARE_BVEC_ITER_ALL(iter_all);
__u16 *expected_guard;
int rc;
total = 0;
- bio_for_each_segment_all(bv, bio, i) {
+ bio_for_each_segment_all(bv, bio, iter_all) {
lnb = iobuf->dr_lnbs[index];
expected_guard = lnb->lnb_guards;
sectors = bv->bv_len / sector_size;
OBD_FREE_PTR(bio_private);
}
-#endif
+#endif /* HAVE_BIO_INTEGRITY_PREP_FN */
+#else /* !CONFIG_BLK_DEV_INTEGRITY */
+#define osd_bio_integrity_handle(osd, bio, iobuf, start_page_idx, \
+ fault_inject, integrity_enabled) 0
+#endif /* CONFIG_BLK_DEV_INTEGRITY */
static int osd_bio_init(struct bio *bio, struct osd_iobuf *iobuf,
bool integrity_enabled, int start_page_idx,
int rc = 0;
bool fault_inject;
bool integrity_enabled;
- DECLARE_PLUG(plug);
+ struct blk_plug plug;
ENTRY;
fault_inject = OBD_FAIL_CHECK(OBD_FAIL_OST_INTEGRITY_FAULT);
unsigned int bi_size = bio_sectors(bio) << 9;
/* Dang! I have to fragment this I/O */
- CDEBUG(D_INODE, "bio++ sz %d vcnt %d(%d) "
- "sectors %d(%d) psg %d(%d) hsg %d(%d)\n",
+ CDEBUG(D_INODE,
+ "bio++ sz %d vcnt %d(%d) sectors %d(%d) psg %d(%d)\n",
bi_size, bio->bi_vcnt, bio->bi_max_vecs,
bio_sectors(bio),
queue_max_sectors(q),
- bio_phys_segments(q, bio),
- queue_max_phys_segments(q),
- 0, queue_max_hw_segments(q));
+ osd_bio_nr_segs(bio),
+ queue_max_segments(q));
rc = osd_bio_integrity_handle(osd, bio,
iobuf, bio_start_page_idx,
fault_inject, integrity_enabled);
}
static int osd_map_remote_to_local(loff_t offset, ssize_t len, int *nrpages,
- struct niobuf_local *lnb)
+ struct niobuf_local *lnb, int maxlnb)
{
+ int rc = 0;
ENTRY;
*nrpages = 0;
int poff = offset & (PAGE_SIZE - 1);
int plen = PAGE_SIZE - poff;
+ if (*nrpages >= maxlnb) {
+ rc = -EOVERFLOW;
+ break;
+ }
+
if (plen > len)
plen = len;
lnb->lnb_file_offset = offset;
lnb->lnb_rc = 0;
lnb->lnb_guard_rpc = 0;
lnb->lnb_guard_disk = 0;
+ lnb->lnb_locked = 0;
LASSERTF(plen <= len, "plen %u, len %lld\n", plen,
(long long) len);
(*nrpages)++;
}
- RETURN(0);
+ RETURN(rc);
}
static struct page *osd_get_page(const struct lu_env *env, struct dt_object *dt,
if (page == NULL)
continue;
- LASSERT(PageLocked(page));
/* if the page isn't cached, then reset uptodate
* to prevent reuse */
if (test_bit(PG_private_2, &page->flags)) {
clear_bit(PG_private_2, &page->flags);
ClearPageUptodate(page);
- unlock_page(page);
+ if (lnb[i].lnb_locked)
+ unlock_page(page);
oti->oti_dio_pages_used--;
} else {
- unlock_page(page);
+ if (lnb[i].lnb_locked)
+ unlock_page(page);
if (pagevec_add(&pvec, page) == 0)
pagevec_release(&pvec);
}
*/
static int osd_bufs_get(const struct lu_env *env, struct dt_object *dt,
loff_t pos, ssize_t len, struct niobuf_local *lnb,
- enum dt_bufs_type rw)
+ int maxlnb, enum dt_bufs_type rw)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
}
}
- osd_map_remote_to_local(pos, len, &npages, lnb);
+ rc = osd_map_remote_to_local(pos, len, &npages, lnb, maxlnb);
+ if (rc)
+ RETURN(rc);
/* this could also try less hard for DT_BUFS_TYPE_READAHEAD pages */
gfp_mask = rw & DT_BUFS_TYPE_LOCAL ? (GFP_NOFS | __GFP_HIGHMEM) :
if (lnb->lnb_page == NULL)
GOTO(cleanup, rc = -ENOMEM);
+ lnb->lnb_locked = 1;
wait_on_page_writeback(lnb->lnb_page);
BUG_ON(PageWriteback(lnb->lnb_page));
return rc;
}
-#ifndef HAVE_LDISKFS_MAP_BLOCKS
-
-#ifdef HAVE_EXT_PBLOCK /* Name changed to ext4_ext_pblock for kernel 2.6.35 */
-#define ldiskfs_ext_pblock(ex) ext_pblock((ex))
-#endif
-
-struct bpointers {
- sector_t *blocks;
- unsigned long start;
- int num;
- int init_num;
- int create;
-};
-
-static long ldiskfs_ext_find_goal(struct inode *inode,
- struct ldiskfs_ext_path *path,
- unsigned long block, int *aflags)
-{
- struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
- unsigned long bg_start;
- unsigned long colour;
- int depth;
-
- if (path) {
- struct ldiskfs_extent *ex;
- depth = path->p_depth;
-
- /* try to predict block placement */
- if ((ex = path[depth].p_ext))
- return ldiskfs_ext_pblock(ex) +
- (block - le32_to_cpu(ex->ee_block));
-
- /* it looks index is empty
- * try to find starting from index itself */
- if (path[depth].p_bh)
- return path[depth].p_bh->b_blocknr;
- }
-
- /* OK. use inode's group */
- bg_start = (ei->i_block_group * LDISKFS_BLOCKS_PER_GROUP(inode->i_sb)) +
- le32_to_cpu(LDISKFS_SB(inode->i_sb)->s_es->s_first_data_block);
- colour = (current->pid % 16) *
- (LDISKFS_BLOCKS_PER_GROUP(inode->i_sb) / 16);
- return bg_start + colour + block;
-}
-
-static unsigned long new_blocks(handle_t *handle, struct inode *inode,
- struct ldiskfs_ext_path *path,
- unsigned long block, unsigned long *count,
- int *err)
-{
- struct ldiskfs_allocation_request ar;
- unsigned long pblock;
- int aflags;
-
- /* find neighbour allocated blocks */
- ar.lleft = block;
- *err = ldiskfs_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
- if (*err)
- return 0;
- ar.lright = block;
- *err = ldiskfs_ext_search_right(inode, path, &ar.lright, &ar.pright);
- if (*err)
- return 0;
-
- /* allocate new block */
- ar.goal = ldiskfs_ext_find_goal(inode, path, block, &aflags);
- ar.inode = inode;
- ar.logical = block;
- ar.len = *count;
- ar.flags = LDISKFS_MB_HINT_DATA;
- pblock = ldiskfs_mb_new_blocks(handle, &ar, err);
- *count = ar.len;
- return pblock;
-}
-
-static int ldiskfs_ext_new_extent_cb(struct inode *inode,
- struct ldiskfs_ext_path *path,
- struct ldiskfs_ext_cache *cex,
-#ifdef HAVE_EXT_PREPARE_CB_EXTENT
- struct ldiskfs_extent *ex,
-#endif
- void *cbdata)
-{
- struct bpointers *bp = cbdata;
- struct ldiskfs_extent nex;
- unsigned long pblock = 0;
- unsigned long tgen;
- int err, i;
- unsigned long count;
- handle_t *handle;
-
-#ifdef LDISKFS_EXT_CACHE_EXTENT /* until kernel 2.6.37 */
- if (cex->ec_type == LDISKFS_EXT_CACHE_EXTENT) {
-#else
- if ((cex->ec_len != 0) && (cex->ec_start != 0)) {
-#endif
- err = EXT_CONTINUE;
- goto map;
- }
-
- if (bp->create == 0) {
- i = 0;
- if (cex->ec_block < bp->start)
- i = bp->start - cex->ec_block;
- if (i >= cex->ec_len)
- CERROR("nothing to do?! i = %d, e_num = %u\n",
- i, cex->ec_len);
- for (; i < cex->ec_len && bp->num; i++) {
- *(bp->blocks) = 0;
- bp->blocks++;
- bp->num--;
- bp->start++;
- }
-
- return EXT_CONTINUE;
- }
-
- tgen = LDISKFS_I(inode)->i_ext_generation;
- count = ldiskfs_ext_calc_credits_for_insert(inode, path);
-
- handle = osd_journal_start(inode, LDISKFS_HT_MISC,
- count + LDISKFS_ALLOC_NEEDED + 1);
- if (IS_ERR(handle)) {
- return PTR_ERR(handle);
- }
-
- if (tgen != LDISKFS_I(inode)->i_ext_generation) {
- /* the tree has changed. so path can be invalid at moment */
- ldiskfs_journal_stop(handle);
- return EXT_REPEAT;
- }
-
- /* In 2.6.32 kernel, ldiskfs_ext_walk_space()'s callback func is not
- * protected by i_data_sem as whole. so we patch it to store
- * generation to path and now verify the tree hasn't changed */
- down_write((&LDISKFS_I(inode)->i_data_sem));
-
- /* validate extent, make sure the extent tree does not changed */
- if (LDISKFS_I(inode)->i_ext_generation != path[0].p_generation) {
- /* cex is invalid, try again */
- up_write(&LDISKFS_I(inode)->i_data_sem);
- ldiskfs_journal_stop(handle);
- return EXT_REPEAT;
- }
-
- count = cex->ec_len;
- pblock = new_blocks(handle, inode, path, cex->ec_block, &count, &err);
- if (!pblock)
- goto out;
- BUG_ON(count > cex->ec_len);
-
- /* insert new extent */
- nex.ee_block = cpu_to_le32(cex->ec_block);
- ldiskfs_ext_store_pblock(&nex, pblock);
- nex.ee_len = cpu_to_le16(count);
- err = ldiskfs_ext_insert_extent(handle, inode, path, &nex, 0);
- if (err) {
- /* free data blocks we just allocated */
- /* not a good idea to call discard here directly,
- * but otherwise we'd need to call it every free() */
- ldiskfs_discard_preallocations(inode);
-#ifdef HAVE_EXT_FREE_BLOCK_WITH_BUFFER_HEAD /* Introduced in 2.6.32-rc7 */
- ldiskfs_free_blocks(handle, inode, NULL,
- ldiskfs_ext_pblock(&nex),
- le16_to_cpu(nex.ee_len), 0);
-#else
- ldiskfs_free_blocks(handle, inode, ldiskfs_ext_pblock(&nex),
- le16_to_cpu(nex.ee_len), 0);
-#endif
- goto out;
- }
-
- /*
- * Putting len of the actual extent we just inserted,
- * we are asking ldiskfs_ext_walk_space() to continue
- * scaning after that block
- */
- cex->ec_len = le16_to_cpu(nex.ee_len);
- cex->ec_start = ldiskfs_ext_pblock(&nex);
- BUG_ON(le16_to_cpu(nex.ee_len) == 0);
- BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
-
-out:
- up_write((&LDISKFS_I(inode)->i_data_sem));
- ldiskfs_journal_stop(handle);
-map:
- if (err >= 0) {
- /* map blocks */
- if (bp->num == 0) {
- CERROR("hmm. why do we find this extent?\n");
- CERROR("initial space: %lu:%u\n",
- bp->start, bp->init_num);
-#ifdef LDISKFS_EXT_CACHE_EXTENT /* until kernel 2.6.37 */
- CERROR("current extent: %u/%u/%llu %d\n",
- cex->ec_block, cex->ec_len,
- (unsigned long long)cex->ec_start,
- cex->ec_type);
-#else
- CERROR("current extent: %u/%u/%llu\n",
- cex->ec_block, cex->ec_len,
- (unsigned long long)cex->ec_start);
-#endif
- }
- i = 0;
- if (cex->ec_block < bp->start)
- i = bp->start - cex->ec_block;
- if (i >= cex->ec_len)
- CERROR("nothing to do?! i = %d, e_num = %u\n",
- i, cex->ec_len);
- for (; i < cex->ec_len && bp->num; i++) {
- *(bp->blocks) = cex->ec_start + i;
- /* unmap any underlying metadata from
- * the block device mapping. b=6998.
- */
- if (pblock != 0)
- clean_bdev_aliases(inode->i_sb->s_bdev,
- *(bp->blocks), 1);
- bp->blocks++;
- bp->num--;
- bp->start++;
- }
- }
- return err;
-}
-
-static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long index,
- int clen, sector_t *blocks, int create)
-{
- int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
- struct bpointers bp;
- int err;
-
- if (index + clen >= inode->i_sb->s_maxbytes >> PAGE_SHIFT)
- return -EFBIG;
-
- bp.blocks = blocks;
- bp.start = index * blocks_per_page;
- bp.init_num = bp.num = clen * blocks_per_page;
- bp.create = create;
-
- CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
- bp.start, bp.start + bp.num - 1, (unsigned)inode->i_ino);
-
- err = ldiskfs_ext_walk_space(inode, bp.start, bp.num,
- ldiskfs_ext_new_extent_cb, &bp);
- ldiskfs_ext_invalidate_cache(inode);
-
- return err;
-}
-
-static int osd_ldiskfs_map_bm_inode_pages(struct inode *inode,
- struct page **page, int pages,
- sector_t *blocks, int create)
-{
- int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
- pgoff_t bitmap_max_page_index;
- sector_t *b;
- int rc = 0, i;
-
- bitmap_max_page_index = LDISKFS_SB(inode->i_sb)->s_bitmap_maxbytes >>
- PAGE_SHIFT;
- for (i = 0, b = blocks; i < pages; i++, page++) {
- if ((*page)->index + 1 >= bitmap_max_page_index) {
- rc = -EFBIG;
- break;
- }
- rc = ldiskfs_map_inode_page(inode, *page, b, create);
- if (rc) {
- CERROR("ino %lu, blk %llu create %d: rc %d\n",
- inode->i_ino,
- (unsigned long long)*b, create, rc);
- break;
- }
- b += blocks_per_page;
- }
- return rc;
-}
-
-static int osd_ldiskfs_map_ext_inode_pages(struct inode *inode,
- struct page **page,
- int pages, sector_t *blocks,
- int create)
-{
- int rc = 0, i = 0, clen = 0;
- struct page *fp = NULL;
-
- CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
- inode->i_ino, pages, (*page)->index);
-
- /* pages are sorted already. so, we just have to find
- * contig. space and process them properly */
- while (i < pages) {
- if (fp == NULL) {
- /* start new extent */
- fp = *page++;
- clen = 1;
- i++;
- continue;
- } else if (fp->index + clen == (*page)->index) {
- /* continue the extent */
- page++;
- clen++;
- i++;
- continue;
- }
-
- /* process found extent */
- rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen,
- blocks, create);
- if (rc)
- GOTO(cleanup, rc);
-
- /* look for next extent */
- fp = NULL;
- blocks += clen * (PAGE_SIZE >> inode->i_blkbits);
- }
-
- if (fp)
- rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen,
- blocks, create);
-
-cleanup:
- return rc;
-}
-
-static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
- int pages, sector_t *blocks,
- int create)
-{
- int rc;
-
- if (LDISKFS_I(inode)->i_flags & LDISKFS_EXTENTS_FL) {
- rc = osd_ldiskfs_map_ext_inode_pages(inode, page, pages,
- blocks, create);
- return rc;
- }
- rc = osd_ldiskfs_map_bm_inode_pages(inode, page, pages, blocks, create);
-
- return rc;
-}
-#else
static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
int pages, sector_t *blocks,
int create)
cleanup:
return rc;
}
-#endif /* HAVE_LDISKFS_MAP_BLOCKS */
static int osd_write_prep(const struct lu_env *env, struct dt_object *dt,
struct niobuf_local *lnb, int npages)
fei.fi_extents_start = &fe;
saved_fs = get_fs();
- set_fs(get_ds());
+ set_fs(KERNEL_DS);
rc = inode->i_op->fiemap(inode, &fei, offset, FIEMAP_MAX_OFFSET-offset);
set_fs(saved_fs);
if (rc != 0)
RETURN(rc);
isize = i_size_read(inode);
- ll_vfs_dq_init(inode);
+ dquot_initialize(inode);
for (i = 0; i < npages; i++) {
if (lnb[i].lnb_rc == -ENOSPC &&
if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW))
SetPageUptodate(lnb[i].lnb_page);
+ if (cache == 0)
+ generic_error_remove_page(inode->i_mapping,
+ lnb[i].lnb_page);
+
if (PageUptodate(lnb[i].lnb_page)) {
cache_hits++;
+ unlock_page(lnb[i].lnb_page);
} else {
cache_misses++;
osd_iobuf_add_page(iobuf, &lnb[i]);
}
-
- if (cache == 0)
- generic_error_remove_page(inode->i_mapping,
- lnb[i].lnb_page);
+ /* no need to unlock in osd_bufs_put(), the sooner page is
+ * unlocked, the earlier another client can access it.
+ * notice real unlock_page() can be called few lines
+ * below after osd_do_bio(). lnb is a per-thread, so it's
+ * fine to have PG_locked and lnb_locked inconsistent here */
+ lnb[i].lnb_locked = 0;
}
end = ktime_get();
timediff = ktime_us_delta(end, start);
lprocfs_counter_add(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
cache_hits + cache_misses);
- if (iobuf->dr_npages) {
+ if (iobuf->dr_npages) {
rc = osd_ldiskfs_map_inode_pages(inode, iobuf->dr_pages,
iobuf->dr_npages,
iobuf->dr_blocks, 0);
- rc = osd_do_bio(osd, inode, iobuf);
+ rc = osd_do_bio(osd, inode, iobuf);
- /* IO stats will be done in osd_bufs_put() */
- }
+ /* IO stats will be done in osd_bufs_put() */
- RETURN(rc);
+ /* early release to let others read data during the bulk */
+ for (i = 0; i < iobuf->dr_npages; i++) {
+ LASSERT(PageLocked(iobuf->dr_pages[i]));
+ unlock_page(iobuf->dr_pages[i]);
+ }
+ }
+
+ RETURN(rc);
}
/*
loff_t diff = i_size_read(inode) - *offs;
spin_unlock(&inode->i_lock);
if (diff < 0) {
- CDEBUG(D_EXT2, "size %llu is too short to read @%llu\n",
+ CDEBUG(D_OTHER,
+ "size %llu is too short to read @%llu\n",
i_size_read(inode), *offs);
return -EBADR;
} else if (diff == 0) {
int size;
int boffs;
int dirty_inode = 0;
+ struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
+ bool create, sparse;
if (write_NUL) {
/*
++bufsize;
}
+ /* sparse checking is racy, but sparse is very rare case, leave as is */
+ sparse = (new_size > 0 && (inode->i_blocks >> (inode->i_blkbits - 9)) <
+ ((new_size - 1) >> inode->i_blkbits) + 1);
+
while (bufsize > 0) {
int credits = handle->h_buffer_credits;
+ bool sync;
+ unsigned long last_block = (new_size == 0) ? 0 :
+ (new_size - 1) >> inode->i_blkbits;
if (bh)
brelse(bh);
block = offset >> inode->i_blkbits;
boffs = offset & (blocksize - 1);
size = min(blocksize - boffs, bufsize);
- bh = __ldiskfs_bread(handle, inode, block, 1);
+ sync = (block > last_block || new_size == 0 || sparse);
+
+ if (sync)
+ down(&ei->i_append_sem);
+
+ bh = __ldiskfs_bread(handle, inode, block, 0);
+
+ if (unlikely(IS_ERR_OR_NULL(bh) && !sync))
+ CWARN("%s: adding bh without locking off %llu (block %lu, "
+ "size %d, offs %llu)\n", inode->i_sb->s_id,
+ offset, block, bufsize, *offs);
+
+ if (IS_ERR_OR_NULL(bh)) {
+ bh = __ldiskfs_bread(handle, inode, block, 1);
+ create = true;
+ } else {
+ if (sync)
+ up(&ei->i_append_sem);
+ create = false;
+ }
if (IS_ERR_OR_NULL(bh)) {
if (bh == NULL) {
err = -EIO;
LASSERTF(boffs + size <= bh->b_size,
"boffs %d size %d bh->b_size %lu\n",
boffs, size, (unsigned long)bh->b_size);
- memcpy(bh->b_data + boffs, buf, size);
+ if (create) {
+ memset(bh->b_data, 0, bh->b_size);
+ if (sync)
+ up(&ei->i_append_sem);
+ }
+ memcpy(bh->b_data + boffs, buf, size);
err = ldiskfs_handle_dirty_metadata(handle, NULL, bh);
if (err)
break;
LASSERT(handle != NULL);
LASSERT(inode != NULL);
- ll_vfs_dq_init(inode);
+ dquot_initialize(inode);
/* XXX: don't check: one declared chunk can be used many times */
/* osd_trans_exec_op(env, handle, OSD_OT_WRITE); */
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(obj));
LASSERT(inode != NULL);
- ll_vfs_dq_init(inode);
+ dquot_initialize(inode);
LASSERT(th);
oh = container_of(th, struct osd_thandle, ot_super);
struct inode *inode = osd_dt_obj(dt)->oo_inode;
u64 len;
int rc;
-
+ mm_segment_t cur_fs;
LASSERT(inode);
if (inode->i_op->fiemap == NULL)
if (fieinfo.fi_flags & FIEMAP_FLAG_SYNC)
filemap_write_and_wait(inode->i_mapping);
+ /* Save previous value address limit */
+ cur_fs = get_fs();
+ /* Set the address limit of the kernel */
+ set_fs(KERNEL_DS);
+
rc = inode->i_op->fiemap(inode, &fieinfo, fm->fm_start, len);
fm->fm_flags = fieinfo.fi_flags;
fm->fm_mapped_extents = fieinfo.fi_extents_mapped;
+ /* Restore the previous address limt */
+ set_fs(cur_fs);
+
return rc;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff